СОВЕРШЕНСТВОВАНИЕ АЛГОРИТМА МАШИННОГО ЗРЕНИЯ В БИОТЕХНИЧЕСКИХ СИСТЕМАХ С ПРИМЕНЕНИЕМ НЕЙРОННЫХ СЕТЕЙ

Abstract

A review of the hardware design of machine vision systems is given in the work; the basic principles of image processing of biotechnical systems and the corresponding mathematical apparatus are considered using the example of the use of machine vision technology in digital dermatoscopy. The digital information extracted by the computer vision system is transferred to special software for image processing using machine learning methods, among which the work highlights the following: neural networks, regression, classification, and object detection. A classification of neural networks that are used for computer vision is given and a brief description of their practical application is given. The considered methods of computer vision systems in biotechnological practice using convolutional neural networks open up opportunities for improving the quality of visual diagnostics. A comparison with previously known methods for classifying digital objects was carried out in the work to analyze the effectiveness of using CNN. Analysis of the digital object by a convolutional neural network using the Statistica software package revealed different color regions. The accuracy of classification of areas of a digital object was 94.7%, which indicates a high accuracy of digital object recognition using CNN and is 1.5% more accurate than statistical classification methods.